[
https://issues.apache.org/jira/browse/MAPREDUCE-6302?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14940517#comment-14940517
]
Karthik Kambatla commented on MAPREDUCE-6302:
---------------------------------------------
Preempting reducers to run mappers doesn't always lead to higher throughput.
The reducer being preempted might have to spend more time to re-copy map
outputs from every mapper than the mappers in question take to run. I
understand that it will likely make sense for the vast majority of cases.
I propose we do the following:
# In this JIRA, let us just fix starvation. Stick to the logic of preempting
enough resources to run one mapper.
# In a follow up JIRA(s), let us improve this preemption to
## preempt reducers until we are able to meet the slowstart threshold
## prioritize preempting reducers that are still in SHUFFLE phase as Jason
mentioned
## add an option to not preempt reducers that are past SHUFFLE phase
irrespective of slowstart as long as one mapper can run
> Incorrect headroom can lead to a deadlock between map and reduce allocations
> -----------------------------------------------------------------------------
>
> Key: MAPREDUCE-6302
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-6302
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Affects Versions: 2.6.0
> Reporter: mai shurong
> Assignee: Karthik Kambatla
> Priority: Critical
> Attachments: AM_log_head100000.txt.gz, AM_log_tail100000.txt.gz,
> log.txt, mr-6302-1.patch, mr-6302-2.patch, mr-6302-3.patch, mr-6302-4.patch,
> mr-6302-prelim.patch, queue_with_max163cores.png, queue_with_max263cores.png,
> queue_with_max333cores.png
>
>
> I submit a big job, which has 500 maps and 350 reduce, to a
> queue(fairscheduler) with 300 max cores. When the big mapreduce job is
> running 100% maps, the 300 reduces have occupied 300 max cores in the queue.
> And then, a map fails and retry, waiting for a core, while the 300 reduces
> are waiting for failed map to finish. So a deadlock occur. As a result, the
> job is blocked, and the later job in the queue cannot run because no
> available cores in the queue.
> I think there is the similar issue for memory of a queue .
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)